The Outlier Lab

Ant International, Mastercard and Visa begin know-your-agent interoperability work

Ant International, Mastercard and Visa began building a shared know-your-agent framework for tracing AI agents to verified operators and users, with no specification, timeline or liability rules yet.

What happened

On September 10, 2026, the three companies announced a collaboration on KYA interoperability, to be run in Singapore through BuildFin.ai, an industry platform convened by the Monetary Authority of Singapore, and to build on MAS's Safeguards for Agentic Finance at Runtime (SAFR) framework. The release names three workstreams:

WorkstreamWhat it aims to do
Cross-network operator traceabilityLink each agent to a validated operator, cardholder or business
Shared certification requirementsAssess agents against agreed security and behavioral requirements
Continuous transaction monitoringKeep evaluating agents using identity and transaction signals

Each network keeps its own verification and decisioning. The work starts from each company's existing protocol: Visa's Trusted Agent Protocol, Mastercard's Verifiable Intent and Ant International's Agentic Mobile Protocol. The stated benefits are fewer duplicate identity checks and lower integration cost for agent platforms and merchants. The release gives no milestones and does not address liability, revocation or disputes.

Mastercard followed on September 30 with trust and intelligence services for its Agent Pay program. The first is a score estimating the probability that an AI agent initiated a transaction, rolling out for testing with US issuers, plus shared signals on whether an agent, merchant, credential or transaction looks unusual. Mastercard also named a partnership with KYA provider Skyfire.

Why it matters for transaction fraud teams

Agent traffic today tends to fall into one of two bad buckets. It looks like a bot and gets blocked by Bot Management and CAPTCHA, or it logs in with the customer's own credentials and looks like Account Takeover (ATO). Operator traceability across networks would give merchants and payment providers a third option, a verified agent lane with its own limits, and give issuers a way to recognize agent-initiated card-not-present spend instead of declining it as anomalous.

The workstreams line up with the chain described in AI Agent Verification (Know Your Agent): who the operator is, who the user is, and ongoing monitoring. The authority question (did the user authorize this purchase?) sits in the underlying protocols, especially Mastercard's intent record, and will matter most in disputes. Until the networks write rules, a "my agent bought it" claim is a Chargeback and Friendly Fraud problem with weak evidence on both sides.

Certification proves that an agent met a bar when it was assessed. It does not stop a certified agent from being steered by injected content, an operator from going rogue, or a real customer from using a real agent to claim every offer in a rewards catalog. Cashback and coupon programs should expect Promo Code Abuse and offer stacking at machine speed from fully verified agents, which makes per-operator and per-user limits the control that matters.

What to do now

  • Add an agent-initiated flag to transaction, refund and dispute records, even if it is only inferred today, so agent losses can be measured apart from human ones (Fraud Labeling and Feedback Loops).
  • Count current agent traffic by type: unsigned automation, signed requests from known operators, agent-specific tokens, and agents using shared customer credentials.
  • Define lanes and limits per verified operator and per delegating user, including caps on offer claims and refunds.
  • Keep your own ability to cut off an operator or key through Negative Lists and Deny Lists and a kill switch, since none of the announced work covers revocation.
  • Ask your acquirer or processor which agent schemes it will support and which verification results it will pass through.
  • Issuers: plan where an agent-initiated probability score would enter authorization and dispute triage, and how you will validate it before relying on it.
  • Decide what evidence you will keep for agent purchases (operator, token, mandate reference, step-up outcome) to match the longest dispute window (Card Network Dispute Rules).

What to watch

  • A technical specification, a participant list and any pilot results from the BuildFin.ai work.
  • Network rule changes that assign dispute liability for agent-initiated transactions.
  • Whether Mastercard's agent probability score moves from testing to general availability, and what issuers report about its accuracy.
  • Adversaries registering as agent operators or impersonating certified agents once a verified lane earns lower friction.

Sources

Updates

  • 2026-10-26: Published.